Sequence-Level Training for Non-Autoregressive Neural Machine Translation

doi:10.1162/COLI_a_00421

	Sequence-Level Training for Non-Autoregressive Neural Machine Translation
	Shao, Chenze 1; Feng, Yang 1; Zhang, Jinchao 2; Meng, Fandong 2; Zhou, Jie 2
	2021-12-01
发表期刊	COMPUTATIONAL LINGUISTICS
ISSN	0891-2017
卷号	47 期号:4 页码:891-925
摘要	In recent years, Neural Machine Translation (NMT) has achieved notable results in various translation tasks. However, the word-by-word generation manner determined by the autoregressive mechanism leads to high translation latency of the NMT and restricts its low-latency applications. Non-Autoregressive Neural Machine Translation (NAT) removes the autoregressive mechanism and achieves significant decoding speedup by generating target words independently and simultaneously. Nevertheless, NAT still takes the word-level cross-entropy loss as the training objective, which is not optimal because the output of NAT cannot be properly evaluated due to the multimodality problem. In this article, we propose using sequence-level training objectives to train NAT models, which evaluate the NAT outputs as a whole and correlates well with the real translation quality. First, we propose training NAT models to optimize sequence-level evaluation metrics (e.g., BLEW based on several novel reinforcement algorithms customized for NAT, which outperform the conventional method by reducing the variance of gradient estimation. Second, we introduce a novel training objective for NAT models, which aims to minimize the Bag-of-N-grams (BoN) difference between the model output and the reference sentence. The BoN training objective is differentiable and can be calculated efficiently without doing any approximations. Finally, we apply a three-stage training strategy to combine these two methods to train the NAT model. We validate our approach on four translation tasks (WMT14 EN <-> De, WMT16 EN <-> Ro), which shows that our approach largely outperforms NAT baselines and achieves remarkable performance on all translation tasks. The source code is available at https://github.com/ictnlp/Seq-NAT.
DOI	10.1162/COLI_a_00421
收录类别	SCI
语种	英语
WOS研究方向	Computer Science ; Linguistics
WOS类目	Computer Science, Artificial Intelligence ; Computer Science, Interdisciplinary Applications ; Linguistics ; Language & Linguistics
WOS记录号	WOS:000753228200006
出版者	MIT PRESS
引用统计	被引频次：14[WOS] [WOS记录] [WOS相关记录]
文献类型	期刊论文
条目标识符	http://119.78.100.204/handle/2XEOYT63/19009
专题	中国科学院计算技术研究所期刊论文_英文
通讯作者	Shao, Chenze
作者单位	1.Chinese Acad Sci, Inst Comp Technol, Key Lab Intelligent Informat Proc, Beijing, Peoples R China 2.Tencent Inc, WeChat AI, Pattern Recognit Ctr, Shenzhen, Peoples R China
推荐引用方式 GB/T 7714	Shao, Chenze,Feng, Yang,Zhang, Jinchao,et al. Sequence-Level Training for Non-Autoregressive Neural Machine Translation[J]. COMPUTATIONAL LINGUISTICS,2021,47(4):891-925.
APA	Shao, Chenze,Feng, Yang,Zhang, Jinchao,Meng, Fandong,&Zhou, Jie.(2021).Sequence-Level Training for Non-Autoregressive Neural Machine Translation.COMPUTATIONAL LINGUISTICS,47(4),891-925.
MLA	Shao, Chenze,et al."Sequence-Level Training for Non-Autoregressive Neural Machine Translation".COMPUTATIONAL LINGUISTICS 47.4(2021):891-925.